Attention Approximates Sparse Distributed Memory
This work offers a theoretical insight into why Attention works well, potentially benefiting researchers in deep learning and neuroscience.
The paper shows that Transformer Attention can be closely related to Sparse Distributed Memory under certain data conditions, which are confirmed in pre-trained GPT2 models, providing new computational and biological interpretations.
While Attention has come to be an important mechanism in deep learning, there remains limited intuition for why it works so well. Here, we show that Transformer Attention can be closely related under certain data conditions to Kanerva's Sparse Distributed Memory (SDM), a biologically plausible associative memory model. We confirm that these conditions are satisfied in pre-trained GPT2 Transformer models. We discuss the implications of the Attention-SDM map and provide new computational and biological interpretations of Attention.